Improve the Frequency Identification in SSVEP based BCI Systems with Moving Windows Algorithm

Abstract

Most processing methods used in SSVEP-based BCI systems use fixed time windows for frequency identification. Due to the variable nature of the EEG signal timing, the use of fixed time windows is not appropriate. In this paper, a new algorithm for floating windows is proposed and evaluated with CCA and LASSO frequency detection methods. The results show that the use of the moving window algorithm for LASSO and CCA methods improves the maximum percentage of frequency identification accuracy by 3.76% and 6.25% respectively. Furthermore, this method decreases the frequency identification time to 0.55 seconds and 0.79 seconds compared to the fixed window algorithm. Advantages such as being able to apply to all frequency recognition methods, increasing the frequency identification accuracy at a certain time of processing compared to fixed windows, adding unlabeled state and adaptability based on system requirements make this algorithm one of the best candidates for SSVEP-based BCI systems.